New insights into the phylogenetics and biogeography of Arum (Araceae): unravelling its evolutionary history
ANAHÍ ESPÍNDOLA
1*, SVEN BUERKI
2, MARIJA BEDALOV
3,4, PHILIPPE KÜPFER
4and NADIR ALVAREZ
51Laboratory of Evolutionary Entomology, Institute of Biology, University of Neuchâtel, Rue Emile-Argand 11, CH-2009 Neuchâtel, Switzerland
2Department of Biodiversity and Conservation, Real Jardín Botánico, CSIC, Plaza de Murillo 2, 28014 Madrid, Spain
3Department of Botany, Faculty of Sciences, University of Zagreb, Marulic´ev Trg 20/II, HR-10000 Zagreb, Croatia
4Laboratory of Evolutionary Botany, Institute of Biology, University of Neuchâtel, Rue Emile-Argand 11, CH-2009 Neuchâtel, Switzerland
5Department of Ecology and Evolution, University of Lausanne, Biophore Building, 1015 Lausanne, Switzerland
The heat- and odour-producing genus Arum (Araceae) has interested scientists for centuries. This long-term interest has allowed a deep knowledge of some complex processes, such as the physiology and dynamics of its characteristic lure-and-trap pollination system, to be built up. However, mainly because of its large distributional range and high degree of morphological variation, species’ limits and relationships are still under discussion. Today, the genus comprises 28 species subdivided into two subgenera, two sections and six subsections. In this study, the phylogeny of the genus is inferred on the basis of four plastid regions, and the evolution of several morphological characters is investigated. Our phylogenetic hypothesis is not in agreement with the current infrageneric classification of the genus and challenges the monophyly of several species. This demonstrates the need for a new infrageneric classification based on characters reflecting the evolution of this enigmatic genus. To investigate the biogeography of Arum deeply, further spatiotemporal analyses were performed, addressing the importance of the Mediterranean basin in the diversification of Arum. Our results suggest that its centre of origin was the European–Aegean region, and that major diversification happened during the last 10 Myr.
ADDITIONAL KEYWORDS: character tracing – infrageneric systematics – Mediterranean biogeography – phylogenetic inferences.
INTRODUCTION
With 109 genera and over 3700 species described (Mayo, Bogner & Boyce, 1997) Araceae have a world- wide distribution and are found in a wide range of environments, from Arctic–Alpine (e.g.Calla palustris L.) to xerophytic (e.g.Anthurium nizandenseMatuda),
with most species occurring in the tropics. The family encompasses a large variety of life forms, from epi- phytic to aquatic, attesting the numerous adaptive radiations that have occurred in this early Cretaceous family (Chaseet al., 2006; Anderson & Janssen, 2009).
A remarkable feature in Araceae is the evolution of heat production in several genera (Minorsky, 2003), especially those displaying pollination-related associa- tions with arthropods, in which thermogenesis is asso- ciated with the emission of volatile compounds and the attraction of pollinators (Moodie, 1975).boj_104914..32
*Corresponding author. E-mail: maria.espindola@unine.ch
One of the few Palaearctic representatives of Araceae is the herbaceous genus Arum L., which comprises 28 described species (Lobin et al., 2007;
CATE project, 2010). Because of its characteristic flowers, showing adaptations for trapping pollinators (Gibernau, Macquart & Przetak, 2004), and its ability to produce heat and odours, this genus has fascinated not only scientists but also the wider public for cen- turies (Theophrastus, 370 BC; Hruby, 1910; Boyce, 1993). As shown by archaeological and historical records, several Arum spp. have been used by humans since ancient times for food (tubers), medi- cine (fruits, leaves, tubers), fashion (tuber starch) and even magic (leaf decoctions) (Prime, 1960).
The inflorescences of Arumconsist of two parts: a spadix and a spathe (Boyce, 1993). The spadix dis- plays the unisexual flowers and harbours adaptations involved in heat production, whereas the spathe is a modified bract surrounding the spadix. One of the distinctive synapomorphies of the genus is the parti- tion of the spadix. The lower zone comprises both female (lower portion) and male (upper portion) flowers placed in a floral chamber, which is usually delimited by male-sterile flowers modified as hairs:
the staminodes. Its apex is a smooth, subcylindrical, usually stipitate organ, known as the spadix appendix (Boyce, 1993). This structure is also recognized as an efficient thermogenetic organ with which the plant attracts pollinating arthropods with heat and produc- tion of volatile compounds. The combination of odour emission and hair presence at the top of the floral chamber (acting as a fence) is a key feature for the efficient trapping of arthropods during the female receptive period and until pollen release (Gibernau et al., 2004).
Historically, the genus was defined by Fuchs (1542) and later established by Linnaeus (1753). The differ- ent species were first circumscribed on the basis of morphology (Schott, 1832), and chromosome counts (Bedalov, 1981) led to the identification of different ploidies in the genus (di-, tetra- and hexaploids, x=14; for a review, see Boyce, 1989). In the most recent revisions of the genus (Boyce, 1993, 1994, 2006; Bedalov & Küpfer, 2005), several morphological characters (tuber shape, flower disposition, growth period, spadix shape and structure of sterile flowers) have been used to build a classification comprising two subgenera, two sections and six subsections. The subgenus Gymnomesium (Schott) Engl. is mono- specific, including only the Hercynian endemicArum pictum L.f. Subgenus Arum Engl. includes sections Arum and Dioscoridea Bronner, the latter being divided into six subsections (Table 1).
This classification may be controversial, notably because: (1) several taxa have been defined on the basis of herbarium specimens (this approach may not
be optimal in this group as important characters are observable only on fresh material; Boyce, 1989); (2) species having large distributions and studied locally were sometimes either simultaneously described under different names (e.g. A. italicum Mill.) or assigned to different taxa when they belonged to the same taxon (e.g. A. cylindraceum Gasp.) (Bedalov &
Küpfer, 2005); (3) following this last point, as several species harbour a high level of intraspecific polymor- phism, this may even trigger the splitting of widely distributed taxa (Boyce, 2006). Therefore, it is now an appropriate time to evaluate the systematics ofArum based on molecular evidence. Published molecular phylogenetic analyses including species ofArumhave focused on the investigation of relationships at the family level and have lacked sampling and resolution at the infrageneric level (Cabrera et al., 2008;
Mansion et al., 2008). In this article, we aim to produce a phylogenetic hypothesis for the genus Arumby sequencing four plastid regions suitable for addressing relationships at the infrageneric level based on 26 of the 28 described species. This will allow us to assess the validity of the current classifi- cation and to examine the evolution of several key characters. To decipher the evolutionary history of this early Miocene genus (Mansion et al., 2008), we perform spatiotemporal analyses to determine the events that played a central role in the radiation, dispersion and isolation of the different species (San- martín, Enghoff & Ronquist, 2001; Médail &
Diadema, 2009). Finally, on the basis of our results, we suggest guidelines for a new infrageneric classifi- cation of the genus Arum.
MATERIAL AND METHODS SAMPLING
Analyses were based on 64 specimens, representing 26 of the 28 described species and spanning all sub- genera, sections and subsections of Arum (Table 1).
On the basis of Mansion et al. (2008), Dracunculus canariensisKunth,D. vulgarisSchott,Biarum davisii Turrill and B. dispar(Schott) Talavera were used as outgroup taxa. Samples were either provided by the DNA Bank of the Royal Botanic Gardens, Kew (UK) or extracted directly from dried plant material from herbaria or field collections (Appendix 1).
DNAEXTRACTION, AMPLIFICATION AND SEQUENCING DNA of freshly collected material and herbarium samples was extracted using the DNeasy Plant Kit (Qiagen, Basle, Switzerland). The plastid regions 3⬘rps16-5⬘trnK, ndhA intron, psbD-trnT and rpl32- trnL were amplified with the primers described in Shawet al. (2007). Amplifications were performed in a
master mix containing 0.5¥buffer, 150 mM deoxy- nucleoside triphosphate (dNTP), 0.7 mM MgCl2, 0.3mgmL-1 bovine serum albumin (BSA), 0.5mM primers and 1 unit of Taq Polymerase (Promega, Dübendorf, Switzerland) made up to a final volume of 30mL with purified MilliQ water. Reactions were run in a TGradient thermocycler (Biometra, Goettingen, Germany). Initial denaturation was programmed for 2 min 30 s at 95 °C, followed by 35 cycles at 95 °C for 35 s, 54–60 °C for 45 s, 72 °C for 1 min and a final extension of 8 min at 72 °C. The purification of PCR products and fluorescence sequencing were performed by Macrogen, Inc. (Seoul, South Korea) and Fasteris Life Sciences (Geneva, Switzerland) with the same primers as used for PCR amplification.
SEQUENCE ALIGNMENT AND PHYLOGENETIC RECONSTRUCTIONS
Automatically generated base-calls for all sequences were checked and edited using ChromasPro 1.41 (Technelysium Pty Ltd, Tewantin, Australia). For each
plastid region, alignment was performed using the ClustalW algorithm implemented in Bioedit 7.0 (Hall, 1999), followed by minor manual corrections. After concatenation of the four regions, a matrix of 3723 bp was obtained. Gaps were further coded following the simple method of Simmons & Ochoterena (2000), as implemented in FastGap 1.2 (Borchsenius, 2009).
The numbers of constant (C), variable (V) and potentially parsimony-informative (PI) sites were cal- culated for each partition using PAUP* v4.0b10 (Swofford, 2002). Before computing total evidence trees, we tested for incongruence among the four regions by applying the partition homogeneity test as implemented in PAUP* v4.0b10 with 100 replicates (this test is equivalent to the incongruence length difference test of Farriset al., 1994; for convenience, it is referred to as the ILD test). Total evidence trees (sensu Kluge, 1989) were determined using both Bayesian inference and maximum parsimony (MP) approaches.
Bayesian Markov chain Monte Carlo (MCMC) methods were used to approximate the posterior prob- Table 1. Current subgeneric taxonomy of genusArumL. Taxa with an ‘*’ were not included in the present study.
Subgenus Section Subsection Species
Gymnomesium A. pictum L.f.
Arum Arum A. byzantinumSchott
A. concinnatumSchott A. italicumMill.
A. maculatumL.
A. megobrebi Lobin, M.Neumann, Bogner & P.C.Boyce
Dioscoridea Alpina A. cylindraceumGasp.
A. lucanum Cavara & Grande Discroochiton A. apulum (Carano) P.Boyce
A. balansanumR.R.Mill A. besserianum Schott A. cyrenaicumHruby A. elongatum Steven A. gratumSchott*
A. hainesii Riedl*
A. nigrumVell.
A. orientaleM.Bieb.
A. purpureospathumP.C.Boyce A. sintenisii P.C.Boyce
Tenuifila A. jacquemontii Blume
A. korolkowiiRegel A. rupicolaBoiss.
Hygrophila A. euxinum R.R.Mill
A. hygrophilumBoiss.
Poeciloporphyrochiton A. dioscoridisSibth. & Sm.
A. palaestinumBoiss.
Cretica A. creticumBoiss. & Heldr.
A. idaeumCoust. & Gand.
ability distribution of the phylogenetic trees on the basis of the combined plastid dataset with four dis- tinct partitions plus one partition for the gap infor- mation, by running MrBayes v.3.1.2 (Ronquist &
Huelsenbeck, 2003). Model selection for the plastid DNA partitions was tested using MrAIC (Nylander, 2004) based on the Akaike information criterion (Akaike, 1973), and a restriction model was applied to the partition containing the coded gaps. Three inde- pendent runs with one cold and five heated chains were run for 5¥107 generations each. Frequencies were sampled every 1000 generations and tempera- ture was fixed to 0.5. The convergence of MCMC was tested by computing the potential scale reduction factor (PSRF; Gelman & Rubin, 1992) as imple- mented in MrBayes, and by determining the effective sample size for each parameter using Tracer v.1.4 (Rambaut & Drummond, 2004). Accordingly, the burn-in period was set to 107 generations until sta- tionarity of the likelihood value was established among the runs, and 10 000 sample points were dis- carded (20% of the total number of trees). The remaining 40 001 trees from each run were pooled (120 003 trees in total) to estimate the posterior prob- ability distribution of the phylogenetic inference. To yield a single phylogenetic hypothesis, the posterior distribution was summarized in the 50% majority- rule consensus tree (referred to as the half-compatible tree in MrBayes), with a Bayesian posterior probabil- ity (BPP) at each node indicating statistical support.
The combined dataset was further analysed under the MP criterion using the parsimony ratchet (Nixon, 1999) as implemented in PAUPrat (Sikes & Lewis, 2001). Based on recommendations by Nixon (1999), ten independent searches were performed with 200 iterations, and 15% of the parsimony-informative characters were perturbed using PAUP* version 4.0b10. The shortest equally most parsimonious trees were combined to produce a strict consensus tree.
Node support was determined by computing decay indices (DIs) (Bremer, 1988) as implemented in TreeRot 3.0 (Sorenson & Franzosa, 2007). DI mea- sures the number of extra steps in tree length required before a node collapses (Bremer, 1988; Baker
& DeSalle, 1997).
Finally, the level of congruence between Bayesian and MP analyses was determined by computing the quartet distance (Estabrook, 1992) between the two topologies. Considering that the distances between the different topologies were small (see Phylogenetic inferences section in Results), the remaining analyses were only based on the Bayesian inference analysis.
CHARACTER EVOLUTION
Character tracing was performed on traits generally used in taxonomic studies ofArum. On the basis of the
topology of the 50% majority-rule Bayesian analysis, the following categorical characters were mapped using Mesquite 2.6 (Maddison & Maddison, 2009) with accelerated transformation optimization (ACCTRAN) and unordered parsimony: tuber form (rhizomatous/
discoid), flower type (flag/cryptic), spadix/spathe ratio (0–0.5; 0.5–1; >1) and ploidy (di-, tetra-, hexaploid).
Characters were obtained from the latest systematic studies performed on the species (Boyce, 1993, 2006;
Bedalov & Küpfer, 2005; Lobinet al., 2007).
DATING AND BIOGEOGRAPHICAL ANALYSES Because the molecular clock hypothesis was rejected (data not shown), the 50% majority-rule Bayesian inference tree was rendered ultrametric using the penalized-likelihood method (Sanderson, 2002; here- after PL), as implemented in the program r8s v.1.71 (Sanderson, 2004) by applying a smoothing value of 1000 and the truncated Newton algorithm. The most external outgroup, B. davisii, was pruned for the estimation of the divergence time as required by the program (see Sanderson, 2004). The following calibra- tion points were applied (according to Mansionet al., 2008): (1) the root node (i.e. the most recent common ancestor of generaArum,BiarumSchott andDracun- culus Mill.) was constrained to a maximum age of 30.2 Mya; (2) the most recent common ancestor of Arum and Dracunculus was constrained to a minimum age of 27.3 Mya; and (3) the stem group of ArumsubgenusArumwas constrained to a minimum age of 16.1 Mya.
Areas were defined following different studies on the geological and biogeographical history of the Mediterranean basin and surrounding areas (Meulen- kamp & Sissingh, 2003; Mansion et al., 2008; Ree &
Sanmartín, 2009), and were set to a number of ten:
East European, West European, Apennines, Aegean, Anatolian, Iranian, Arabian, North African, Macaro- nesian Islands and Caucasus (Fig. 3). The rules applied to define the area for each species were as follows: (1) if the origin of the sample was known, the sample was attributed to the area in which it was sampled; (2) if the origin was unknown, the sample was assigned to the area in which the plant is known to be distributed according to Boyce (1993, 2006) and the search engine ofFlora Europaea(Flora Europaea, 2009) [in the case of A. balansanum R.R.Mill., A.
byzantinumSchott,A. sintenisii(Engl.) P.C.Boyce,D.
vulgaris,D. canariensisandB. dispar]; (3) if a sample belonged to a widely distributed and well-described species for which we did not possess samples from all the parts of the distributional area, it was assigned to its region of origin plus the remaining noncovered regions according to Boyce (1993) (only in the case of A. italicum).
Dispersal-vicariance analysis (DIVA) is a method for inferring the most parsimonious reconstruction of ancestral ranges on a given phylogenetic tree by minimizing the number of dispersal and extinction events that are needed to explain the current termi- nal distributions (Ronquist, 1997). The program DIVA (Ronquist, 2001) uses a three-dimensional cost matrix to estimate the cost of moving from the ances- tor to each of the descendants (Ronquist, 1997). It allows two different scenarios for range inheritance at speciation nodes: (1) duplication or within-area speciation, when the ancestor is distributed in a single area and each of the two descendants inherits the entire ancestral range (e.g. A to A); (2) vicariance, when the ancestor occurs in two or more areas and each descendant inherits a nonoverlapping subset of the ancestral range (e.g. AB to A and B). Only one dispersal event per branch (between two ancestral nodes) is allowed in the model, except for terminal branches leading to widespread taxa, for which DIVA postulates multiple dispersal events. To account for polytomies in the 50% majority-rule Bayesian infer- ence tree, five more exceptions were required in our analysis (see below). DIVAs were run with the maximum number of areas allowed at ancestral nodes constrained to two. Uncertainty in phyloge- netic relationships was accounted for in DIVA by using an approach proposed by Nylander et al.
(2008), which integrates DIVA parsimony-based reconstructions over a Bayesian MCMC sample of trees representing the posterior probability of the tree topology (hereafter referred to as Bayes-DIVA).
Specifically, we sampled one tree for every 16 trees (7501 in total) from the MCMC ‘post-burnin’ sample and used R scripts available from the second author to summarize/average ancestral area reconstructions over all sampled trees for each node in the 50%
majority-rule Bayesian inference, which was used as the reference. Only those trees containing the node of interest were summarized in estimating the prob- abilities for that node. This approach allows an esti- mation of the marginal probabilities of ancestral ranges for a given node whilst integrating over the uncertainty in the rest of the tree topology (Nylander
et al., 2008). Ancestral areas and vicariance/dispersal events were recorded following Buerki (2009).
As several polytomies were found in the 50%
majority-rule topology, the following rule was applied to solve incompatibilities between nodes and to esti- mate correct dispersal-vicariance events (that other- wise would violate DIVA assumptions): if the most probable area for a given node was incompatible (according to DIVA assumptions) with that of the next coming node or tip, it was combined with the follow- ing most probable area(s); this was performed until the ancestral areas of the node were congruent with the areas assigned to the following node or tip. In order to summarize the different dispersal events across the three geological epochs spanning the diver- sification of Arum (Miocene, Pliocene and Pleis- tocene), a pairwise matrix of dispersion was built for each epoch to address the links among the ten defined areas [this was performed using R (R Development Core Team, 2009), with scripts available on request from the second author]. When a branch spanned over more than one epoch, the proportion of the branch over each epoch was considered, and the fraction D of one single dispersal event in a given epoch was recorded (0<D<1). To summarize the results, arrows with variable widths (proportional to the number of dispersal events) were drawn on palaeo- geographical maps corresponding to the three rel- evant epochs (Meulenkamp & Sissingh, 2003).
RESULTS
PHYLOGENETIC INFERENCES
The combined dataset consisted of 250 sequences.
Aligned lengths were 845 bp for 3⬘rps16-5⬘trnK, 1077 bp for the ndhA intron, 1024 bp for psbD-trnT, 777 bp for rpl32-trnL and 104 binary positions corresponding to coded gaps. The final matrix thus contained a total of 3827 characters (3723 nucleotides and 104 gap presence/absence). Values for C, V and PI for each partition are given in Table 2. Partition rpl32-trnL provides slightly more informative sites (36) than the other partitions. The partition homoge- Table 2. Sequenced regions, with corresponding total number and percentages of base pairs (bp), constant (C), variable (V) and parsimony informative (PI) sites.
Region Total (bp)
Constant (C) sites
Variable (V) sites
Parsimony-informative (PI) sites
3⬘rps16-5⬘trnK 845 (100%) 791 (93.6%) 54 (6.4%) 17 (2.0%)
ndhAintron 1077 (100%) 1021 (94.8%) 56 (5.2%) 24 (2.2%)
psbD-trnT 1024 (100%) 993 (97.0%) 31 (3.0%) 13 (1.3%)
rpl32-trnL 777 (100%) 730 (94.0%) 47 (6.0%) 36 (4.6%)
neity test was passed (P=0.07), indicating that the information provided by the four plastid regions was congruent. Topologies obtained with Bayesian infer- ence and MP algorithms (Fig. 1) were highly con- gruent (quartet distance of 0.114, meaning that approximately 89% of the components were compat- ible between the two trees) and defined five major supported clades (see Fig. 1). Topologies depict A.
pictum as sister to the rest of the genus, confirming the definition of subgeneraGymnomesium andArum (supported by BPP=1 and DI=10; Fig. 1). In subge- nus Arum, the first branching clade I includes A.
palaestinumBoiss. and the different subspecies ofA.
dioscoridis Sibth. & Sm. (BPP=1, DI=3). Clade II containsA. concinnatum Schott andA. italicum (for which monophyly was not contradicted in the Baye- sian analysis) (BPP=0.95, DI=3). Clade III includes the two most eastern taxa, A. jacquemontii Blume andA. korolkowiiRegel,A. megobrebiLobin, M.Neu- mann, Bogner & P.C.Boyce,A. rupicolaBoiss. and the two easternmost samples ofA. maculatumL. included in this study, and a new species, hereafter referred to asA.sp. nov. (BPP=0.99, DI=1). In the MP topology, A. creticum Boiss. & Heldr. is also included in clade III as the first branching lineage (Fig. 1). Clade IV includes two subclades containing A. sintenisii P.C.Boyce and A. hygrophilum Boiss., on the one hand, andA. byzantinumSchott,A. nigrumVell. and some specimens ofA. elongatumSteven, on the other (BPP=0.84, DI=1). In the Bayesian inference topol- ogy, A. idaeum Coust. & Gand. is also included in clade IV as the first branching lineage (Fig. 1).
Finally, clade V is poorly resolved and includes the remaining taxa:A. maculatum(western samples), A.
cylindraceum Gasp.,A. orientale M.Bieb., A. besseri- anum Schott, A. balansanum R.R.Mill, A. pur- pureospathum P.C.Boyce, A. euxinum R.R.Mill, A.
apulum(Carano) P.Boyce, A. cyrenaicumHruby and one representative ofA. elongatum(BPP=1, DI=2).
The relative position of clade V swapped depending on the phylogenetic algorithm, as it was sister to clade IV in the Bayesian inference tree, but sister to clade III in the MP tree (Fig. 1). Incongruence between the two topologies concerned (1) the positions of clade V relative to clades III and IV, and (2) the branching ofA. idaeum andA. creticum.
Sections as defined by Boyce (1989) were not sup- ported by the phylogenetic hypotheses. Except for the cases of subsectionsPoeciloporphyrochiton(clade I,A.
dioscoridis and A. palaestinum) and Tenuifila (sub- clade in clade III,A. jacquemontii,A. korolkowiiand A. rupicola), our topologies did not support the current infrageneric delimitation (Fig. 1, Table 1).
Finally, the monophyly of several widespread species (e.g. A. elongatum and A. maculatum) was not sup- ported (see Fig. 1).
CHARACTER EVOLUTION
The reconstruction of ancestral states for the four studied characters is shown in Figure 2. The trait that appears to be most constrained from the phylo- genetic reconstruction is ploidy (Fig. 2A), with an ancestral character state corresponding to diploidy (2n=2x=28) and one single evolution towards hexaploidy. Tetraploidy evolved several times. The remaining characters (Fig. 2B, tuber shape; Fig. 2C, flower shape; Fig. 2D, spathe/spadix ratio) show a pattern of multiple independent events and are much less informative at the infrageneric level. An excep- tion could be the evolution of the rhizomatous tuber shape which, although largely symplesiomorphic, seems to be correlated with the level of ploidy.
BIOGEOGRAPHICAL ANALYSIS
Reconstructed ancestral areas for the different nodes corresponding to the 50% majority-rule Bayesian tree are shown in Figure 3. The two most probable ances- tral areas from the crown nodes of the genus were the Aegean and West European regions. Later nodes show that the Aegean and Anatolian regions were the only areas to harbour ancestral lineages of the genus for a long time. Overall, a substantial proportion of the dispersion ofArumlineages towards their current distribution areas seems to have happened after the late Miocene (c. 10 Mya).
The rates and direction of dispersal events at three different time-slices corresponding to the Miocene, Pliocene and Pleistocene are shown in Figure 4.
During the Miocene (23–5.3 Mya, Fig. 4A), dispersion mainly happened from the Aegean area to Anatolia.
Exchanges were also possible between the newly emerging Caucasus region and the Aegean and Ana- tolian areas. Colonization of the Iranian area seems to have happened only during this period. Coloniza- tion of Macaronesia also occurred at this time, but other dispersals towards this area were probably also possible during the early Pliocene (5.3–2.6 Mya, Fig. 4B). During this epoch, the genus extended its distribution for the first time onto the Arabian plate, and important dispersion events seem to have happened from the Aegean (and, to a lesser extent, from the Anatolian region) to Eastern Europe. The North African region was colonized during the most recent geological epoch (Pleistocene, 2.6 Mya–present, Fig. 4C) probably via two pathways: (1) from the Apennines through the southern tip of the Italian Peninsula; and (2) from the Arabian region through the Gulf of Suez. Numerous dispersals also occurred from the Aegean to the Anatolian area in the Pleis- tocene and from the Apennine region to Western and Eastern Europe and the Aegean. During this last epoch, exchanges seemed to have halted between the
A
Clade VClade IClade IIClade IIIClade IV
A. apulum A. balansanum 1
A. besserianum A. orientale 2 A. cylindraceum 3
A. maculatum 2 A. maculatum 3
A. maculatum 1
A. maculatum 9
A. maculatum 6 A. maculatum 7 A. maculatum 8
A. cylindraceum 4 A. cylindraceum 5 A. cylindraceum 6 A. cylindraceum 7
A. cylindraceum 8 A. cylindraceum 1
A. cylindraceum 9 A. cylindraceum 2
A. lucanum 2
A. lucanum 1 A. cyrenaicum A. euxinum
A. purpureospathum
A. balansanum 2 A. elongatum 3
A. balansanum 3 A. orientale longispathum A. orientale 1
A. byzantinum A. elongatum 2 A. elongatum 1 A. nigrum A. hygrophilum
A. sintenisii A. idaeum
A. creticum
A. jacquemontii A. korolkowii A. rupicola rupicola A. rupicola virescens A. maculatum 4
A. maculatum 5 A. megobrebi 1 A. megobrebi 2
A. megobrebi 3 A. megobrebi 4
A. sp. nov A. concinnatum
A. italicum canariense 1 A. italicum canariense 2 A. italicum albispathum A. italicum italicum
A. dioscoridis 2 A. dioscoridis cyprium A. dioscoridis 1 A. palaestinum A. pictum
Dracunculus canariense
Biarum dispar
Biarum davisii
Dracunculus vulgaris2 Dracunculus vulgaris1
Figure 1. Inferred plastid phylogenies: A, Bayesian inference, half-compatible tree; B, maximum parsimony (MP), strict consensus tree. Values shown on the branches represent Bayesian posterior probability values (A) and decay indices (B).
Vertical bars indicate major clades explained in the text.
B
3 8
16
4 10
3 1
2
3 2 1 4
2
1 5
1 1
1
1 1 7
4 1
2 1 1 2
2 A. apulum A. balansanum 1 A. besserianum
A. orientale 2
A. cylindraceum 3 A. maculatum 2 A. maculatum 3
A. maculatum 1
A. maculatum 9
A. maculatum 6 A. maculatum 7
A. maculatum 8 A. cylindraceum 4 A. cylindraceum 5 A. cylindraceum 6
A. cylindraceum 7
A. cylindraceum 8 A. cylindraceum 1
A. cylindraceum 9 A. cylindraceum 2
A. lucanum 2
A. lucanum 1 A. cyrenaicum A. euxinum
A. purpureospathum
A. balansanum 2 A. elongatum 3 A. balansanum 3 A. orientale longispathum
A. orientale 1
A. byzantinum A. idaeum A. creticum A. jacquemontii
A. korolkowii A. rupicola rupicola
A. rupicola virescens A. maculatum 4 A. maculatum 5 A. megobrebi 1 A. megobrebi 2 A. megobrebi 3
A. megobrebi 4 A. sp . nov
A. elongatum 2 A. elongatum A. nigrum
A. hygrophilum A. sintenisii
A. concinnatum A. italicum canariense 1 A. italicum canariense 2
A. italicum albispathum A. italicum italicum A. dioscoridis 2 A. dioscoridis cyprium
A. dioscoridis 1 A. palaestinum
A. pictum Dracunculus canariense
Biarum dispar Biarum davisii
Dracunculus vulgaris2 Dracunculus vulgaris1
9.0
Figure 1. Continued
A. apulum A.balansanum1 A. euxinum A.cyrenaicum A. purpureospathum A. lucanum 1 A. cylindraceum 1 A.cylindraceum 7 A. cylindraceum 6 A. cylindraceum 5 A. cylindraceum 4 A. cylindraceum 8 A. maculatum 9 A. cylindraceum 3 A.maculatum 2 A.maculatum 3 A. lucanum 2 A. maculatum 1 A. besserianum A. orientale 2 A. cylindraceum 2 A. maculatum 6 A. maculatum 7 A.maculatum 8 A. balansanum 2 A. elongatum 3 A. balansanum 3 A. orientale longispat A. orientale 1 A. cylindraceum 9 A. byzantinum A. elongatum 2 A. elongatum 1 A. nigrum A. hygrophilum A. sintenisii A. idaeum A. creticum A. jacquemontii A. korolkowii A. rupicola rupicola A. rupicola virescens A. maculatum 4 A. maculatum 5 A. megobrebi 1 A. megobrebi 2 A. megobrebi 3 A. megobrebi 4 A. sp nov A. concinnatum A. italicum albispathum A. italicum A. italicum canariense1 A. italicum canariense2 A. dioscoridis 2 A. dioscoridis cyprium A. dioscoridis 1 A. palaestinum A. pictum Drac. canariense Drac. vulgaris1 Drac. vulgaris2 B. davisii B. dispar
Ploidy Level diploid hexaploid
Tuber shape discoid rhizomatous
A B
tetraploid Figure2.EvolutionofcategoricalcharactersontheBayesiantopology:A,ploidy(2x,4x,6x);B,tubershape(rhizomatous,discoid);C,flowertype(flag,cryptic); D,spadix/spatheratio(0–0.5;0.5–1;>1).Coloursareexplainedinthelegendofeachfigure.Missingsquaresatthelevelofterminaltaxaindicateunknown characters.
Flower type cryptic flag
Ratio spadix/spathe <0.5 >1
C D
0.5 - 1
A. apulum A.balansanum1 A. euxinum A.cyrenaicum A. purpureospathum A. lucanum 1 A. cylindraceum 1 A.cylindraceum 7 A. cylindraceum 6 A. cylindraceum 5 A. cylindraceum 4 A. cylindraceum 8 A. maculatum 9 A. cylindraceum 3 A.maculatum 2 A.maculatum 3 A. lucanum 2 A. maculatum 1 A. besserianum A. orientale 2 A. cylindraceum 2 A. maculatum 6 A. maculatum 7 A.maculatum 8 A. balansanum 2 A. elongatum 3 A. balansanum 3 A. orientale longispat A. orientale 1 A. cylindraceum 9 A. byzantinum A. elongatum 2 A. elongatum 1 A. nigrum A. hygrophilum A. sintenisii A. idaeum A. creticum A. jacquemontii A. korolkowii A. rupicola rupicola A. rupicola virescens A. maculatum 4 A. maculatum 5 A. megobrebi 1 A. megobrebi 2 A. megobrebi 3 A. megobrebi 4 A. sp nov A. concinnatum A. italicum albispathum A. italicum A. italicum canariense1 A. italicum canariense2 A. dioscoridis 2 A. dioscoridis cyprium A. dioscoridis 1 A. palaestinum A. pictum Drac. canariense Drac. vulgaris1 Drac. vulgaris2 B. davisii B. dispar Figure2.Continued
Biarum dispar
Pleistocene
A. apulum A. balansanum 1
A. besseranium A. orientale 2
A. cylindraceum 3 A. maculatum 2 A. maculatum 3 A. cylindraceum 4 A. cylindraceum 5 A. cylindraceum 6 A. cylindraceum 7
A. lucanum 2 A. maculatum 1 A. cylindraceum 8 A. cylindraceum 1 A. cyrenaicum A.euxinum A. lucanum 1 A.maculatum 9 A. purpureospathum A. cylindraceum 2 A. maculatum 6 A. maculatum 7 A. maculatum 8 A. balansanum 2 A. elongatum 3 A. balansanum 3 A. orientale longispathum A. orientale 1 A. cylindraceum 9 A. byzantinum A. elongatum 2 A. elongatum 1 A. nigrum A. hygrophilum A. sintenisii A. idaeum A. creticum
A. jacquemontii A. korolkowii A. rupicola rupicola A. rupicola virescens A. maculatum 4 A. maculatum 5
A. megobrebi 1 A. megobrebi 2 A. megobrebi 3 A. megobrebi 4 A. sp. nov.
A. concinnatum A. italicum albispathum A. italicum
A. italicum canariense 1 A. italicum canariense 2 A. dioscoridis 2
A. dioscoridis cyprium A. dioscoridis 1 A. palaestinum A. pictum Drac. canariense Drac. vulgaris1 Drac. vulgaris2
Aegean and Eastern Europe, despite the neighbour- ing position of these two regions.
DISCUSSION
INFRAGENERIC RELATIONSHIPS AND SPECIES’IDENTITY
Classical taxon definition and circumscription in the genusArum(Boyce, 1993, 2006) only partially match our phylogenetic hypothesis. As shown in Figure 1, the identity of sections, subsections (sensu Boyce, 1989, 1993) and species is strongly challenged, and it seems obvious that there is a ‘gap’ between the current classification and the genetic identities of the taxa. Our analyses, however, confirm the validity of the two subgenera, asA. pictum(subgenusGymnomesium) is the first branching lineage ofArum(as shown previ- ously by Mansionet al., 2008). This result is supported by floral (staminodes present, but no pistillodes) and phenological (flowering in autumn and not in spring as in the rest of the species) characters restricted toA.
pictum. For more than a century, the peculiarities of this Hercynian endemic have been recognized, and several authors have attempted to place it in a differ- ent genus (Gymnomesium Schott, 1855). The long branch separating this lineage from the other species (with a divergence estimated to be sometime between the early and middle Miocene; Fig. 3) favours the hypothesis of a palaeorelictual identity ofA. pictum(as proposed by Mansion et al., 2008). Within subgenus Arum, only subsection Poeciloporphyrochiton (Fig. 1, clade I) is retrieved by our phylogenetic hypothesis. It appears as the first branching lineage in the subgenus, confirming thatA. dioscoridisandA. palaestinumare closely related and placed in an external position, as proposed formerly by several authors (Hruby, 1910;
Boyce, 1989, 1993). Another exception could be subsec- tion Tenuifila, which is nested in clade III in the Bayesian topology, and might still be considered as a valid entity (see below). No other subsection is com- patible with our results.
Both Bayesian and MP topologies argue in favour of the monophyly of hexaploid taxa (Fig. 1, clade II;
Fig. 2A), with A. italicum specimens clustering together with A. concinnatum. The insular A. itali- cum ssp. canariense (Webb & Berthel.) P.Boyce is genetically differentiated from the ‘continental’ sub-
species from which it diverged during the Pliocene (Fig. 3). As the monophyly of A. italicum is not retrieved in the MP analysis, a more thorough analy- sis (e.g. using genomic screening markers) should be performed to confirm the status ofA. concinnatum.
Although the phylogenetic relationships among the three remaining clades (III, IV and V) are not yet resolved (i.e. the topology varies according to the phylogenetic algorithm), their respective monophyly is relatively well supported with DIⱖ1 and BPP>0.95 (with the exception of clade IV, which shows a lower support of 0.84; see Fig. 1). Current molecular data do not allow the discussion of the phylogenetic relation- ships ofA. creticumand A. idaeum, two species with overlapping distributions in Crete, which are either placed in a polytomy at the base of these three clades or as the first branching lineages of clade III (MP topology) and clade IV (Bayesian topology), respec- tively. These species are morphologically similar (open floral chambers, sweet or weak odour production vs.
closed floral chamber and clear lure-oriented odour production in the other species) and were included in subsectionCretica(Boyce, 1989). Relationships among clades III, IV and V should also be examined carefully as our results do not allow conclusions to be drawn regarding the position of clade V, as it appears as either sister to clade III (MP topology) or to clade IV (Baye- sian inference tree).
The strongly supported clade III (excludingA. creti- cum; BPP=0.99; DI=1) comprises all members of subsection Tenuifila(i.e.A. rupicola,A. jacquemontii andA. korolkowii, which form a well-supported mono- phyletic group in the Bayesian topology; BPP=0.98) and all representatives of A. megobrebi, two speci- mens ofA. maculatumfrom the easternmost edge of the distribution and one sample from the Caucasus area. The latter should be considered as a new species (referred to asA.sp. nov.). The placement ofA. macu- latum samples in clade III is unexpected as the two specimens found here are highly divergent phyloge- netically with respect to the European representa- tives (clade V) of this widely distributed species (Fig. 1). As a consequence, they might merit treat- ment as a different species if further morphological studies confirm this status by identifying synapomor- phies. Clade III has a biogeographical coherence as the taxa included are found in the eastern part of the Figure 3. Ancestral areas assigned by dispersal-vicariance analysis (DIVA) to each node of the Bayesian topology.
Colours represent ancestral areas (see legend). All areas with a probability<0.1 were pooled and treated as a single undetermined area (in black). Nodes having been treated with special rules are indicated by ‘*’ or ‘#’: ‘*’ indicates that the second most probable area has been combined with the first in order to sketch compatible scenarios; ‘#’ indicates that more than two ancestral areas have been combined to obtain the congruence of the nodes and tips. Scale corresponds to million years from present. Map shows areas defined for the biogeographical inference.
䉳
Miocene (23Mya - 5.3 Mya)
Pliocene (5.3Mya - 2.6Mya)
West-European
East-European
Apennine Aegean
Anatolian
Caucasus
Iranian
Arabian North-African
Macaronesian
West-European
East-European
Apennine
Aegean
Anatolian
Caucasus
Iranian
Arabian North-African
Macaronesian
A
B
distribution area of the genus (from northern Turkey to the Iranian region), confirming the role played by this area in the diversification ofArum(as proposed previously by Bedalov & Küpfer, 2005).
Clade IV (excludingA. idaeum; BPP=0.84; DI=1) comprises two specimens of A. elongatum, which appears to be paraphyletic with respect toA. nigrum, and possibly A. byzantinum (in the Bayesian topol- ogy), although the position of the latter is not well supported (Fig. 1). More generally, the morphology typical for A. elongatum seems to be quite labile as one specimen is also found in clade V (see below). The other representatives of this clade are the closely related A. sintenisii (endemic to Cyprus) and the orientalA. hygrophilum, the former probably having diverged from the latter in the late Pliocene after a dispersal followed by an insular differentiation. This result was already predicted by Boyce (2006).
Finally, clade V is by far the least resolved, encom- passing closely related taxa that diverged during the second half of the Pliocene and the Pleistocene (Fig. 3), most having colonized the Apennines and temperate habitats in Western and Eastern Europe (Fig. 4). In this clade, differentiation among speci- mens is weak and all species sampled more than once are paraphyletic (Fig. 1). Uncovering the relation-
ships among taxa within this clade would require further analyses based on, for example, genomic screening. This might help to address the status of widely distributed taxa, such asA. cylindraceumand the ‘European’ A. maculatum, with that of narrow endemics such as A. apulum and A. pur- pureospathum. Another case of interest is the well- supported group composed ofA. besserianumand one specimen of A. orientale (BPP=1; DI=2; Fig. 1).
Arum orientale is still poorly defined as attested by successive revisions during the last 15 years (Boyce, 1994, 2006; Bedalov & Küpfer, 2005). This taxon was first described as a species with several subspecies, present in Crimea and extending to the eastern part of the Balkans (Boyce, 1993). However, several mor- phological characters point to a close relationship with A. besserianum distributed in Ukraine and Poland (P. Küpfer, pers. observ.). Consequently, this taxon certainly encompasses different paraphyletic lineages, and both its status and that of other taxa (e.g. A. balansanum) should be investigated using novel genomic techniques coupled with taxonomy.
Therefore, we have observed that two different pat- terns arise when testing the monophyly of species for which more than one specimen was collected. On the one hand, some are well supported by our analyses:
Pleistocene (2.6Mya - present)
West-European
East-European
Apennine
Aegean
Anatolian Caucasus
Iranian
Arabian North-African
Macaronesian
C
Figure 4. Dispersion events at three time slices: A, Miocene; B, Pliocene; C, Pleistocene (maps A and B, modified from Meulenkamp & Sissingh, 2003; with permission of the editors). The widths of arrows are proportional to dispersal rates.
Broken lines indicate boundaries between biogeographical zones.
this is the case forA. dioscoridis(clade I),A. rupicola (clade III) and, to a lesser extent,A. megobrebi(clade III) (Fig. 1). On the other hand, some species are clearly polyphyletic (comprising specimens from lin- eages that diverged as early as the Pliocene), such as, for instance,A. maculatum(clades III and V) andA.
elongatum(clades IV and V) (Fig. 1). Finally, the case ofA. italicum(clade II) is somewhat intermediate, as the monophyly is not contradicted by the Bayesian topology, whereas the species is paraphyletic (with the inclusion of A. concinnatum) in the MP topology.
CHARACTER EVOLUTION:CHARACTERIZING THE IDENTITY OF THE MOST RECENT COMMON ANCESTOR Among all the investigated characters, only ploidy seems to be related to the evolution of the genus (Fig. 2A). The remaining traits (spathe/spadix ratio, flower type, shape of the tuber) show patterns of multiple independent evolution and a looser correla- tion with the evolutionary history of Arum. Our results therefore support the hypothesis that the level of ploidy might constitute an informative character for the systematics of the genus, as first proposed by Bedalov & Küpfer (2005), and address the diploid status of the most recent common ancestor (a trait shared by both A. pictum and taxa within clade I).
However, the abrupt transition from diploidy to hexaploidy (in clade II) seems to be unlikely and might require the existence of a transitional and yet extinct or undiscovered tetraploid form. The advan- tages of polyploids in terms of survival have been addressed recently in Arum, as artificial crossings between distinct species yielded polyploid hybrids that were ‘robust and maintain themselves in culti- vation without apparent difficulties’ (Bedalov &
Küpfer, 2005). Interestingly, the fact that clade V comprises both a substantial proportion of tetraploid lineages and an important number of recognized taxa could support the idea of an increased fitness in polyploids, facilitative for the radiation of this group (Fig. 3) (for a review of the ability of polyploids to colonize a wider range of habitats, see Prentis et al., 2008). However, before arriving at any conclusion, it is important that the phylogenetic relationships among the specimens of this clade are clarified.
One local phylogenetic constraint on the flower type (flag vs. cryptic) was addressed in the subclade cor- responding to theTenuifila subsection, with all taxa sharing a flag flower, whereas the ancestral state for this trait withinArumwas a cryptic flower. As there is a strong association between cryptic flowers and attract-and-lure pollination strategies (Boyce, 1989;
Gibernauet al., 2004), the latter should be considered as the ancestral pollination mode in the genus. It is important, however, to mention that the polyphyletic
status of this character is not surprising, as it is related to reproductive structures, which, in Arum, appear to be highly correlated with fast-evolving pol- lination syndromes (Chouteau, Gibernau & Barabé, 2008). The evolution of this character would thus reflect more strongly the ecological processes that species have independently undergone rather than the evolutionary history of the genus.
Finally, our results confirm that the ancestral state of theArumtuber shape was discoid, as proposed by Bedalov & Küpfer (2005), and that the appearance of the first rhizomatous species happened at the same time as the transition from diploidy to hexaploidy (clade II, Fig. 1). Although the transition from a discoid to a rhizomatous tuber occurred several times, there seems to be a trend towards a correlation between ploidy and tuber shape: all rhizomatous species are polyploid. In contrast, several polyploid species (A. apulum, A. cyrenaicum and A. pur- pureospathum) have discoid tubers. As sections withinArumwere classically defined on the basis of this homoplasious character, there is a strong need to consider morphological characters from other plant parts to build a new classification ofArumcompatible with our molecular evidence.
ARUM THROUGH SPACE AND TIME
Bayes-DIVA provides strong support for an Aegean/
Western European origin of the genus (Fig. 4) some- time in the early Miocene (c. 20 Mya). However, assuming that the earliest diverging lineage (now composed of onlyA. pictum) originated and survived in the Hercynian islands long before all otherArumspp.
arose (according to the palaeorelictual hypothesis pro- posed by Mansion et al., 2008), the ancestral area corresponding to the rest of the genus is the Aegean region (Fig. 3). This zone has acted as a natural laboratory allowing the diversification of lineages, sometime in the late Miocene (Figs. 3, 4). The Aegean also appears to be a main source of dispersal events throughout the evolutionary history of Arum. Its central position with respect to the other areas in which the genus is present today could have facilitated this. Most dispersal events recorded during the middle to late Miocene occurred from the Aegean to the Anatolian region (Fig. 4A). Later, the emergence of the Iranian plate allowed its colonization once a land- bridge was established with the Anatolian plate in the late Miocene (Meulenkampet al., 2000). During this period, no dispersals were observed towards the Arabian plate that was still isolated from the northern lands by a marine transgression (Meulenkamp &
Sissingh, 2003). Once the Caucasian archipelago emerged (and possibly after its uplift and contact with the Northern Anatolian region), further dispersals to
and from this region occurred in the late Miocene. At this same period, long-distance dispersals are recorded from the Aegean to the Macaronesian regions.
The first dispersal to the Arabian zone occurred more recently, during the Pliocene, when a sea regres- sion (Meulenkamp & Sissingh, 2003) allowed this land to come into contact with the Anatolian region (Fig. 4B). The regression of the western Para-Tethys (following the uplift of the European plates) could also have permitted the dispersal from the Aegean to Eastern Europe. Exchanges continued between the Aegean and Anatolia and through these two zones to the Caucasus and Eastern Europe.
During more recent times (Pleistocene), new dis- persals from the Aegean to Anatolia were recorded (Fig. 4C), probably facilitated by the Mediterranean regressions characterizing Quaternary climatic oscil- lations (Peulvast et al., 2000). At this time, North Africa was colonized twice through the Arabian plate and through the Apennines (Fig. 4C).
Although exchanges mainly occurred longitudinally (east–west) across land paths at the periphery of the seas during the early history of the genus, the pace of dispersion and diversification inArumincreased after the peri-Mediterranean region was unified (i.e. after the emergence of the Arabian and Iranian plates, the uplift of the Caucasus and the regressions of Tethys and Para-Tethys) (Meulenkampet al., 2000; Meulen- kamp & Sissingh, 2003).
The colonization of Macaronesia requires special treatment. The taxon inhabiting this region (A. itali- cum ssp. canariense) appears to have arrived there during the late Miocene or early Pliocene, in agree- ment with the timing of colonization already observed in several other endemic taxa of these islands (Carine et al., 2004). This ancient dispersal contrasts with the more recent colonization of North Africa. This discor- dance could be a result of either a first colonization of North Africa, having allowed the dispersal towards Macaronesia through mid-distance dispersal and further extinction of this lineage, or a long-distance dispersal directly from the Northern Peri-Tethys.
Considering the morphology of the seeds (Mayoet al., 1997), this latter hypothesis could be possible only in association with animals; birds have already been proposed as the main dispersers of A. maculatum (Snow & Snow, 1988), which could also be true forA.
italicum(Méndez, 1997).
TOWARDS A NEW CLASSIFICATION OFARUM? Because our phylogenetic reconstruction strongly con- tradicts the current systematics of the genus, the need for a new classification is evident (i.e. a large number of homoplasies are suggested by the tracing of the characters currently used in the delimitation of
sections and species in our topology; Fig. 2). However, we recommend caution in formally proposing a new infrageneric classification until nonmolecular synapo- morphies supporting the main clades are identified.
The two current subgenera, Arum and Gymnome- sium, are supported by our analyses, although the status of the latter might be reconsidered, given the high level of phylogenetic differentiation of this mono- specific subgenus. The characteristic morphology, development, distribution and, as shown in this study, phylogenetic position displayed by A. pictum could indicate that it would be more correct to place it in the monospecific genus Gymnomesium (as formerly proposed by Schott, 1855).
The two formerly defined sections within subgenus Arumare not supported by the phylogenetic analyses and, based on molecular evidence, we recommend a division of the subgenus into five sections (corre- sponding to clades I–V). Subsection Poeciloporphyo- chiton(corresponding to clade I) should be elevated to the rank of section, whereas new synapomorphies should be recovered for the other clades. The sectional classification ofA. creticumandA. idaeumshould also be investigated more thoroughly as our phylogenetic hypotheses only weakly associate them with clades III and IV, respectively. Although this study demon- strates the importance of ploidy as a putative syna- pomorphy in the case of clade II, a broad survey of morphological characters is strongly recommended.
Finally, the paraphyletic status of widespread species requires additional analyses to be performed with more variable markers in order to validate these findings. Nonetheless, our results already argue for a revision of species such as A. maculatum and A.
elongatumin which major splits have been identified.
Future taxonomic revisions should carefully consider characters not related to pollination, as lineages seem to be able to adapt quickly to changes in pollinator availability, leading to floral character convergence in distinct clades.
ACKNOWLEDGEMENTS
The authors would like to thank the different persons and institutions that kindly provided the material used in this study: the DNA Bank of the Royal Botanic Gardens, Kew (UK), as well as Dr Cusimano (University of Münich, Germany), Dr Lobim and Mr Neumann (Botanical Garden, University of Bonn, Germany). Discussions with Dr Bogner (Botanical Garden Munich, Germany) also largely contributed to the design of the manuscript. This study was funded by the Swiss National Science Foundation (project no.
3100A0-116778) and the Swiss Academy of Sciences (SCNAT+).
